Issues

• Layerwise HF export silently ignores save_modelopt_state=True

  bug
  NVIDIA/Model-Optimizer#3040 · Filter by author yingguo-trt opened on Oct 27, 2026

• INT4 packing extension never loads from a read-only install

  NVIDIA/Model-Optimizer#3025 · Filter by author ConnorBaker opened on Oct 26, 2026

• [Puzzletron] Runtime stats benchmark model computes head size as hidden_size // num_attention_heads instead of reading head_dim

  bug
  NVIDIA/Model-Optimizer#3016 · Filter by author kevalmorabia97 opened on Oct 26, 2026
  

• XiaomiMiMo/MiMo-V2.6-Flash-RL & XiaomiMiMo/MiMo-V2.6-Pro-RL nvfp4

  feature request
  NVIDIA/Model-Optimizer#2865 · Filter by author functionstackx opened on Oct 17, 2026

• GGML fake-quant keeps stale codes after a bitsandbytes optimizer step, so QAT/QAD silently stops training the quantized weights

  NVIDIA/Model-Optimizer#2857 · Filter by author christopherrobertbrooks-tech opened on Oct 17, 2026

• fp4_kernel_hopper.py uses tl.make_block_ptr, which is removed in Triton main (breaks NVFP4 fake quant)

  NVIDIA/Model-Optimizer#2810 · Filter by author apbose opened on Oct 14, 2026
  

• Support activation-only scale search for PyTorch AWQ calibration

  NVIDIA/Model-Optimizer#2674 · Filter by author thuiaaa opened on Oct 6, 2026
  

• warnings.warn(f"RealQuantLinear: No real-quant GEMM found: {self}.")

  bug
  NVIDIA/Model-Optimizer#2593 · Filter by author zewenli98 opened on Sep 29, 2026
  3

  

• [Bug] qdq_to_dq cannot convert per-tensor quantized weights (scalar scale raises IndexError, (1,) scale raises shape mismatch)

  NVIDIA/Model-Optimizer#2561 · Filter by author SID-6921 opened on Sep 27, 2026
  3

  

• [Bug] modelopt-mcp: all launcher subprocesses hang when the server runs as a stdio MCP server — missing stdin=DEVNULL

  NVIDIA/Model-Optimizer#2551 · Filter by author lucky20260806 opened on Sep 26, 2026
  1

  

• Your project is on StackMap — a curated map of the AI stack

  NVIDIA/Model-Optimizer#2481 · Filter by author hoghweed opened on Sep 19, 2026

• calling mtq.quantize causes AttributeError: 'tuple' object has no attribute 'dim'

  question
  NVIDIA/Model-Optimizer#2424 · Filter by author ccarmatic opened on Sep 13, 2026
  3

  

• Only 1 sample from AutoCast calibration data set is truly used

  bug
  NVIDIA/Model-Optimizer#2337 · Filter by author lashgar opened on Sep 4, 2026
  3

  

• After applying NVFP4 quantization followed by compression, RealQuantLinear fails to locate the corresponding GEMM implementation during inference.

  bug
  NVIDIA/Model-Optimizer#2330 · Filter by author Lee-YNU opened on Sep 4, 2026
  5

  

• ONNX PTQ: no NVFP4 path for convolutional models

  NVIDIA/Model-Optimizer#2281 · Filter by author geoffrey-delhomme opened on Aug 28, 2026
  1

  

• [Feature Request] DeepSeek-V4-Flash-0731 NVFP4 checkpoint

  feature request
  NVIDIA/Model-Optimizer#2220 · Filter by author nicole-lihui opened on Aug 20, 2026
  2

• LUT-B Support

  question
  NVIDIA/Model-Optimizer#2204 · Filter by author brian-dellabetta opened on Aug 17, 2026
  2

  

• [Bug] init_quantized_weights / --low_memory_mode exports numerically broken NVFP4 (quantizes meta tensors before weights load)

  NVIDIA/Model-Optimizer#2160 · Filter by author spped2000 opened on Aug 12, 2026
  1

  

• Can auto_quant calculate the score for kv-cache separately?

  question
  NVIDIA/Model-Optimizer#2158 · Filter by author zcfh opened on Aug 12, 2026
  1

  

• Support MiniMax-H3 Visual VAE quantization in ModelOpt

  feature request
  NVIDIA/Model-Optimizer#2137 · Filter by author baonudesifeizhai opened on Aug 10, 2026
  1

  

• # [ONNX][Autotune] Integrated quantization does not preserve AutoTune Q/DQ placement on ViT

  bug
  NVIDIA/Model-Optimizer#2123 · Filter by author HannisLee opened on Aug 10, 2026
  2

  

• [ONNX PTQ] Support for third-party custom ORT/TRT plugins in static calibration

  feature request
  NVIDIA/Model-Optimizer#2016 · Filter by author e-said opened on Jul 24, 2026
  8

  

• Recommended FP8 recipe for Blackwell (B200)? Per-tensor FP8_DEFAULT_CFG underperforms BF16 at prefill; NVFP4 much faster. Also: exported FP8 checkpoint has no KV q/k/v_scale

  bug
  NVIDIA/Model-Optimizer#2015 · Filter by author Francesco-Carlucci opened on Jul 24, 2026
  5

  

• NVFP4 support for Qwen3_5MoeExperts (fused MoE): quantizer registration + export serializer

  NVIDIA/Model-Optimizer#2011 · Filter by author frischeDaten opened on Jul 23, 2026
  10

  

• Public preset quant_cfg changed from dict to list in 0.44 (breaking, undocumented?)

  NVIDIA/Model-Optimizer#2001 · Filter by author sayakpaul opened on Jul 21, 2026
  8